PyRIT
microsoft.github.io
- Free plan
- Linux
- Web
- API
- Self-hosted
What PyRIT is
PyRIT is a free, MIT-licensed open-source framework for security professionals and engineers assessing risks in generative AI systems. It supports single-turn and multi-turn red-team strategies, including Crescendo, TAP, and Skeleton Key, alongside repeatable scenario evaluations across hundreds of objectives such as content harms, psychosocial risks, and data leakage. CoPyRIT provides a web interface for interacting with AI systems, tracking findings, and collaborating with a team. Documented targets include OpenAI, Azure, Anthropic, Google, HuggingFace, custom HTTP endpoints and WebSockets, and web apps tested with Playwright. PyRIT can store conversations, scores, and attack results in SQLite or Azure SQL, and export results. Scoring options include true/false, Likert scale, classification, and custom scorers, including LLM-based and Azure AI Content Safety scoring. The project documents Scanner, GUI, and Framework modes, with command-line assessment through pyrit_scan and pyrit_shell. Installation options include Docker or local installation with pip or uv; local installation documentation recommends Python 3.13. The project directs vulnerability reports to the Microsoft Security Response Center rather than public GitHub issues.
Who it suits
PyRIT suits security professionals and engineers who assess generative AI systems. Its scenario evaluations and team interface also support repeatable, collaborative assessment work.
Good to know, and what to watch
- Free and MIT licensed
- Includes single-turn and multi-turn attack strategies
- Evaluates hundreds of scenario objectives
- Supports a web interface for collaborative findings
- Offers several storage and scoring options
- Local installation documentation recommends Python 3.13
- Support is limited to listed resources
- Vulnerabilities should not be reported through public GitHub issues
Plans
Every plan and pricePyRIT
Open source
The verdict
PyRIT offers multiple assessment modes, targets, and scoring approaches for teams evaluating generative AI risks. Users should note the stated support limits and vulnerability-reporting route.
What the maker publishes
- Human-led interface
- CoPyRIT provides a web interface for interacting with AI systems, tracking findings, and collaborating with a team. source
- Installation
- The project documents user installation through Docker or locally with pip or uv, plus contributor Docker and local installation options. source
- License
- The GitHub repository identifies PyRIT as MIT licensed. source
- Memory
- PyRIT can store conversations, scores, and attack results in SQLite or Azure SQL, and supports exporting results. source
- Purpose
- PyRIT is an open-source framework for security professionals and engineers to identify risks in generative AI systems. source
- Python compatibility
- The current documentation recommends Python 3.13 for local installation. source
- Red teaming
- It supports single-turn and multi-turn attack strategies, including Crescendo, TAP, and Skeleton Key. source
- Scenario evaluations
- Its scenario framework supports repeatable evaluations across hundreds of objectives, including content harms, psychosocial risks, and data leakage. source
- Scoring
- Response scoring supports true/false, Likert scale, classification, and custom scorers, including LLM-based and Azure AI Content Safety scoring. source
- Security reporting
- The documentation directs vulnerability reports to the Microsoft Security Response Center and says not to report them through public GitHub issues. source
- Support
- The project uses GitHub Issues for bugs, feature requests, and questions, and says support is limited to the resources it lists. source
- Targets
- Documented targets include OpenAI, Azure, Anthropic, Google, HuggingFace, custom HTTP endpoints and WebSockets, and web apps tested with Playwright. source
- Ways to use
- The documentation describes Scanner, GUI, and Framework modes, including command-line assessment through pyrit_scan and pyrit_shell. source
Checked against the maker's pages on 8 Oct 2026.